Upload
truongthu
View
221
Download
0
Embed Size (px)
Citation preview
-1-
Context Effects for Census Measures of Race and Hispanic Origin
Elizabeth Martin, Theresa J. DeMaio, and Pamela C. Campanelli
Center for Survey Methods Research
U. S. Bureau of the Census
ABSTRACT
This paper reports on the results of a split-ballot experiment conducted
in 1987 to test alternative versions of the decennial census long form. Two
forms were randomly assigned and self-administered in group sessions involving
a total of 515 respondents. The order of race and Hispanic origin items was
experimentally manipulated. The standard long form asks race, then Hispanic
origin. The experimental form reversed the order of the items in order to
reduce perceived redundancy, and to create a more restricted frame of
reference for the race item. The objectives of the context manipulation were
(1) to reduce item nonresponse for the Hispanic origin item, and (2) to reduce
reporting of "Other race" by Hispanics in the race item. Objective (1) was
met. Objective (2) was met for Hispanics born in a U. S. State, but not for
immigrants. The results are interpreted as reflecting a process of
acculturation which affects how Hispanic respondents apply U. S. racial
categories "White" and "Black" in the census.
-2-
The survey methodology literature contains much evidence of the effect of
question sequencing on response to questions. This work has, in
large part, involved attitude items (see Schuman and Presser, 1981; Turner
and Martin, 1984). The effect of question order may derive from the context
invoked by prior questions, which may influence respondents' frame of
reference or suggest differing interpretations of the question.
Context effects can also operate on quasi-factual and factual items, but
research in this area has been limited. The potential for this type of effect
is especially pronounced when the concept being measured is somewhat unclear,
and the respondent really isn't sure what is being asked. In this paper, we
explore the effect of question sequence on responses to two potentially
ambiguous items on the decennial census questionnaire: race and Hispanic
origin.
The Measurement of Race
Despite its familiarity, the concept of race is not a simple one. Racial
classifications, both popular and scientific, are based on a mixture of
principles and criteria, including national origin, tribal membership,
religion, language, minority status, physical characteristics, and behavior.
The criteria and categories for racial classification vary among cultures and
over time. In the United States, we are accustomed to think in terms of two
major races: Black and White.
In this country, we tend to treat race as an objective, fixed
characteristic of a person which is biologically inherited. This meaning of
race is so ingrained that it may come as a surprise to learn that other
cultures have very different conceptions of race (Marshall, 1968; Harris,
1968). For example, the racial categories recognized in Brazil are not the
-3-
same as those used in the United States, even though its population also
includes components with White European and Black African origins. According
to Harris (1968), numerous racial categories are combined to create hundreds
of different racial terms, based primarily on physical characteristics.
Racial descent is not the rule; full siblings whose appearance differs are of
different races. In addition, race is not a fixed characteristic of a person,
and it changes when a person achieves wealth, for example, which is one of the
criteria for race.
Cultures in Central and South America use different racial categories and
criteria from those used in the United States. This difference in racial
classifications implies that "White" and "Black" are not natural categories of
race for most Spanish-speaking people. This fact has implications for the
consistency and meaningfulness of the answers Spanish-speaking immigrants may
give to the census race question.
Even within the United States, there have been significant shifts in how
Americans understand and categorize race, which have been reflected in changes
in the race categories used in the census.
Race was first measured as a separate item in the 1850 census, using the
categories, "White," "Black," and "Mulatto," which were also used in 1860. In
the 1870 and 1880 censuses, categories for "Chinese" and "Indian" were added,
and in 1890, "Japanese" was added. The interpretation of race as a biological
(yet observable) characteristic in these early censuses is clear. Enumerators
were instructed in 1870 to code race through observation and to, "Be
particularly careful in reporting the class Mulatto. The word here is
generic, and includes quadroons, octoroons, and all persons having any
perceptible trace of African blood. Important scientific results depend upon
the correct determination of this class . . . ." (U.S. Bureau of the Census,
1979:18). The 1890 census saw an elaboration of the "Mulatto" category, with
separate categories for mulattoes, quadroons, and octoroons. In 1900, the
-4-
attempt to distinguish mixed Black and White categories was abandoned but
taken up again in 1910, when "Mulatto" was again listed as a category. In
1930, "Mulatto" was dropped permanently, and "A person of mixed White and
Negro blood was to be returned as Negro, no matter how small the percentage of
Negro blood" (U.S. Bureau of the Census, 1979:52).
There were also changes in the census racial classification of Spanish-
speaking persons, such as Mexicans. Enumerators in 1930 were instructed that
the race of "all persons born in Mexico, or having parents born in Mexico, who
were not definitely White, Negro, Indian, Chinese, or Japanese, would be
returned as Mexicans" (U.S. Bureau of the Census, 1979:52). In 1940, this
rule was changed, and Mexicans and other persons of "Latin descent" were to be
classified as "White" unless they were definitely of some race other than
White. The rule was changed again in 1980, and Hispanic entries, such as
Puerto Rican and Mexican, were left in the "Other race" category.
Finally, there were additions to the Asian categories. In 1960,
categories were added for Filipino, Hawaiian, Part Hawaiian, Aleut, and Eskimo
(Alaska and Hawaii had become states in 1959). Eskimo and Aleut were dropped
in 1970 but restored in 1980, by which time there were 9 separate Asian and
Pacific Islander categories, including Asian Indian, Guamanian, Samoan, and
Vietnamese.
Since we think of race as a stable, enduring characteristic, it is
fascinating to find that no single set of racial categories has been used in
more than two censuses, and most were only used once. The changes partly
reflect real change in the composition of the U.S. population due to migration
from Asia and Central and South America, and expansion of U.S. territory to
include new groups. Undoubtedly, political considerations and pressures
played a part as well in the decisions to create and combine categories.
The fluctuations in census race categories also suggest the difficulty of
devising a meaningful, objective classification of race. Marshall (1968)
-5-
argues that all racial classifications are arbitrary and artificial to some
extent, whether they are based on "scientific" or popular criteria. Problems
arise when respondents do not share the race categories used by the Census
Bureau. Evidence suggests this is the case for many Spanish-speaking persons.
In 1970, the Census Bureau began conducting the census by self-enumeration
rather than personal visit enumeration. Personal enumeration was conducted
only for households that did not mail back a form. The change in census-
taking procedure meant that, after 1970, race was based mainly on self-
identification rather than enumerator observation.1
In conjunction with the change in procedure, dramatic changes in the
racial characteristics of the population were observed. The number of
Hispanic origin persons classified as "Other race" rose from 700,000 in 1970
to 5.8 million in 1980 (U. S. Bureau of the Census, 1987a:100). The
transition to self-enumeration contributed to this increase and the decline in
"White" race reporting among people of Hispanic origin. In addition, in 1980
persons of Hispanic origin were no longer recoded as "White" if they reported
themselves as "Other race," as had been done in the 1970 census.
Additional evidence links self-enumeration with higher "Other race"
reporting by Hispanics. In 1980, 35 percent of Hispanics were identified as
"Other race" in census questionnaires, but only 10 percent were reported as
"Other race" in personal visit reinterviews conducted after the census
(McKenney, Fernandez, and Masamura, 1985). A possible reason for the
inconsistency is respondent confusion with the 1980 census race question (U.S.
Bureau of the Census, 1987a:100), which did not explicitly ask for race. Many
persons filled in the circle for "Other race" and wrote in a nationality from
Central or South America. Another possible cause is interviewer behavior in
the reinterview study. Race was to be self-reported by respondents using a
flashcard listing the race categories. Possibly, interviewers changed "Other
race" responses to "White" for respondents who "looked White." Or, possibly
-6-
Hispanic respondents gave different answers in a personal interview (quite
probably conducted by a White interviewer) than they gave on the census
questionnaire.
The Measurement of Hispanic Origin
The Hispanic origin item was included in the census for the first time in
1970. In 1970, the item appeared only on the long form (a more detailed and
extensive set of questions that goes to one in six households), but in 1980 it
was asked of everyone as a 100-percent item. Placement near the race item on
the questionnaire in 1980 may have affected reporting for both items.
Hispanic origin had the highest nonresponse rate of any 100-percent item
in 1980. The computer allocation rate was 4.2 percent for short forms and 2.3
percent for long forms (U.S. Bureau of the Census, 1986: 32). The item may
have appeared redundant to respondents who had just reported "Other race" and
written in a Hispanic nationality. Many non-Hispanic persons may have left
the item blank because they didn't understand it, or because they did not find
a category that fit them.
Evidence from the 1970 and 1980 censuses suggests lack of understanding of
the question. Siegel and Passel (1979) and Passel and Word (1987) document
extensive misreporting in 1970 in the "Central or South American" category by
persons living in the central and southern United States. In 1980, fairly
substantial numbers of people (particularly Blacks in the South, Northeast,
and Midwest) provided false positive reports of Mexican-American origin (U. S.
Bureau of the Census, 1982). Respondents apparently wanted to indicate that
they were American, and the Mexican-American response category was the only
one that contained the word "American" (Passel and Word, 1987).
Some Hispanics also were confused by the question. A pattern of
inconsistent reporting of "Other Hispanic" origin was observed between the
1980 census and the Content Reinterview Study (U.S. Bureau of the Census,
1986; McKenney, Fernandez and Masamura, 1985). While reporting by Puerto
-7-
Ricans, Mexicans, and Cubans was fairly consistent (that is, the census agreed
with the reinterview for 85 percent or more of the persons in the reinterview
sample), only 55 percent of persons reported as "Other Hispanic" in the
reinterview had been similarly classified in the census; the others had been
reported mostly as Mexican (11 percent) and as non-Hispanic (31 percent).
Fernandez and Cresce (1986) found inconsistent reporting of Hispanic
origin and Hispanic ancestry in the 1980 census. Many persons reported
Hispanic origin but not Hispanic ancestry (e.g., 30 percent reported as Puerto
Rican in origin did not report Puerto Rican ancestry). There were also over a
half million reports of Hispanic ancestry in the census by persons who did not
report Hispanic origin. The inconsistency may be evidence of the part-whole
contrast effect described in the survey methodology literature (Schuman and
Presser, 1981). This occurs when respondents "subtract out" their response to
a previous question when answering a subsequent, more general one.
Respondents to the ancestry question had already reported their Hispanic
origin; when faced with a very similar question, they may have reinterpreted
it to be non-redundant: "Besides what you already told me about, what is your
ancestry?" Thus, these seemingly inconsistent reports may be evidence of
respondents' attempts to make sense of the questions and provide non-redundant
information.
Additional support for this interpretation is provided by an experiment
which varied the contiguity of the race and Hispanic origin items (see
McKenney, Cresce and Johnson, 1988; U.S. Bureau of the Census, 1987b). The
Hispanic origin item nonresponse rates ranged from 20 to 30 percent when
placed immediately after race, compared to 13 to 17 percent when placed three
items after race. This suggests that, with consecutive sequencing, the race
item sets a frame of reference for the Hispanic origin item which accentuates
the perceived redundancy of the latter item.
Hypotheses
-8-
We hypothesized that the high nonresponse for the Hispanic origin question
and the reporting of "Other" in the race question both may result from context
effects. The item that comes first may create a frame of reference that
affects interpretation of the second item. The abundance of national origin
groups listed as categories for race may encourage write-in entries of other
nationality groups in that item. The Hispanic origin item then appears
redundant, leading respondents to leave it blank (see figures 1 and 2).
INSERT FIG. 1 AND 2 HERE
We reasoned that the Hispanic origin item is less redundant if it precedes
race. People who think their answer to the Hispanic origin question can be
inferred from their response to the race question would be less likely to
leave the Hispanic origin question blank if they answer it first. By
reversing the order of the two items, we hoped to decrease item nonresponse
for Hispanic origin.
The reporting of "Other" in the race item may also be affected by the
order of the two items. The majority of "Other" races written in are Hispanic
nationalities. We thought that giving Hispanics a chance to report their
Hispanic origin first would create a more restricted frame of reference for
the race item, and reduce their reporting of "Other race." However, if
Hispanics reject the categories "White" and "Black," then their reporting of
"Other" race may be insensitive to the order in which the items are asked.
(See DeMaio and Martin, 1987, for a statement of the hypotheses on which our
revisions to the census form were based.)
Methodology
This study was conducted as part of a program of research to improve the
design of the decennial census long form questionnaire. In this redesign
effort, typographic and other layout changes were made to increase the
consistency of the form's appearance, clearer instructions were provided on
how to complete the census form, questions were reworded to simplify and
-9-
clarify concepts, and questions were reordered to improve the flow and
coherence of the census form (see DeMaio, Martin, and Sigman, 1987; Bates and
DeMaio, 1989, for further discussion of the goals and methods of the
research).
Figures 1 and 2 contain facsimiles of the race and Hispanic origin items
for the revised form and the 1986 test census form, which was used as a
control. As can be seen, several changes were made to these items, the
principal one being that the items were reversed to test the hypotheses
discussed above. In both forms, race and Hispanic origin were separated by
other items (age in the revised form; age and marital status in the 1986
form). Other revisions were also made. Response categories for the Hispanic
origin item were reordered, as they were throughout the revised form, so that
"Yes" preceded "No." An instruction to fill in the "No" circle if the person
was not Hispanic was added. There were also differences in response
categories in the revised form, which reflected the thinking within the Census
Bureau at the time regarding the content of these items in 1990. In the 1986
form, detailed Hispanic (for Hispanic origin) and Asian (for race) categories
were listed separately. In the revised form, categories were combined, with
an instruction to write in each person's specific group.
The revised and 1986 forms were compared in a series of split-panel
experiments conducted in April 1987 in 33 group sessions organized by the
Census Bureau Regional Offices in Boston, Dallas, Chicago and Philadelphia.
Participants in the sessions were volunteers recruited by Regional Office
outreach staff through their contacts with community groups. Recruiting was
focused on minority racial and ethnic groups, and people with relatively
little education. Among the volunteers who appeared for the sessions, there
were a total of 9 nonrespondents who did not attempt to fill out census forms
(3 refusals, 2 persons with visual impairments, 2 with language problems, and
2 late arrivals). A total of 515 people filled out forms for themselves and
-10-
members of their households, providing data on 1,446 persons. Participants
included people aged 18 to 80, members of different racial and ethnic groups,
and people with various levels of education. (See Parsley, et al., 1989, for
a detailed report of the results of this research.)
Each session lasted an hour and a half, with 55 minutes allotted to fill
out the census long form. (Forty-five minutes is the Census Bureau's official
estimate of the time it takes an average household to fill out this form.)
During each session, half of the participants were randomly assigned the 1986
form and the other half were assigned the revised form. Even though these
respondents do not represent a probability sample, the randomization by form
type permits us to make statistical comparisons between forms. We tested for
statistically significant form differences using chi-square tests that take
into account the clustering of persons within group sessions and within
households (Fay, 1989). In the tables that follow, the chi-square values
labelled X are Pearson chi-squares, calculated on the assumption of simple2
random sampling; the J's refer to jack-knife statistics that take into account
clustering in the data and thus represent a more conservative test.
Likelihood ratio chi-squares (identified as L in the tables) were used to2
test the fit of alternative log-linear models.
Results
Form Differences in Distributions of Race and Hispanic Origin: Despite
revisions to the race question and response categories, race distributions are
very similar for the two forms. As shown in the first panel of Table 1, each
form identified about the same percentage of individuals in the categories of
White, Black, Asian or Pacific Islander, and Other race. The revised form,
however, identified a significantly larger percentage of American Indians.
This result is due to sampling variability and is not a form effect.2
INSERT TABLE 1 HERE
The second panel of Table 1 contains the distribution of responses to the
-11-
Hispanic origin item, collapsed into the categories on the revised form for
comparability. Again, the two forms obtained similar levels of reporting of
Hispanic origin. Although based on small numbers, this result is encouraging
because it suggests that the increased response to the item in the revised
form (as reported below) did not alter the distribution.
Item Nonresponse: Results in Table 2 support our hypothesis concerning item
nonresponse to the Hispanic origin item. When Hispanic origin is asked first,
item nonresponse drops to 9 percent, compared to 18 percent on the 1986 form.
Nonresponse for the race item was not affected by the change in order, with a
rate of 3 percent for the 1986 form and 4 percent for the revised form.
INSERT TABLE 2 HERE
Because of the number of changes made to the form, we cannot be sure that
the reversed order is the cause of the better response rate for Hispanic
origin in the revised form. However, the pattern of missing data by race
leads us to believe that the sequence of the items and the additional
instruction to fill in the "No" circle were responsible. Results for the 1986
form (see the first panel of Table 3) show that persons who reported their
race as "Black" or "Asian/Pacific Islander" were far more likely to leave the
Hispanic origin item blank than persons reported as "White" or "Other race".
On the revised form, however, all race groups were equally likely to leave the
Hispanic origin item blank (see second panel of Table 3).
INSERT TABLE 3 HERE
A loglinear model confirms the presence of a three-way interaction
(Nonresponse to Hispanic origin X Race X Type of form). This finding supports
our hypothesis that the race item conditions respondents' understanding of the
intent of the Hispanic origin item. Our interpretation is that Blacks and
Asians are especially likely to think their response ("No") can be inferred
from their response to the race item, and hence they leave Hispanic origin
blank. When Hispanic origin is first and there is an explicit instruction to
-12-
fill in the "No" circle for persons who are not Hispanic, persons of all races
are likely to give a response.
Content Differences--Race: Our second major hypothesis concerned the level of
reporting of "Other race" by persons of Hispanic origin. For both
questionnaire versions, the vast majority of write-in entries in "Other race"
were Hispanic (over 90 percent for both forms). This finding is consistent
with prior evidence.
The results suggest that the form of the questionnaire did affect Hispanic
individuals' responses to the race item. As shown in Table 4, persons who
reported Hispanic origin were more likely to report their race as "White" in
the revised form (39 percent) than in the 1986 form (25 percent) although this
difference is not quite statistically significant with the more conservative
test (p=.12). However, the extent of reporting "Other race" write-ins by
Hispanics was substantial on both forms (61 percent for the revised form vs.
75 percent for the 1986 form). Thus, our goal of reducing reporting of "Other
race" by Hispanics was only partially successful. These data suggest that
most of the persons of Hispanic origin who participated in our test did not
identify with any of the race categories listed in the census questionnaire.
INSERT TABLE 4 HERE
Further analysis shows that the effect of context was restricted to
Hispanics who were born in the United States. Table 5 presents responses to
the race item, separately for people born in a U. S. State or outside the
United States (i.e., in a foreign country or U.S. territory); the table
includes data for persons of Hispanic origin only, and excludes those for whom
information on place of birth is missing. The first panel shows that race
reporting for Hispanics who were not born in a U. S. State is unaffected by
the version of the questionnaire. For this group, race is reported as "Other"
and an Hispanic nationality is written in for over three-quarters of the cases
on both forms. However, there is a very large context effect on race
-13-
reporting for Hispanic persons who were born in a U. S. State. For U. S.-born
Hispanics, race was significantly more likely to be reported as "White" in the
revised form (74 percent) than in the 1986 form (22 percent). (The three-way
interaction between form, place of birth, and race is marginally significant.)
INSERT TABLE 5 HERE
This result suggests that most Hispanic immigrants simply do not apply the
categories "White" or "Black" to themselves, regardless of whether or not they
have first reported their Hispanic origin. Race reporting for U. S.-born
Hispanics, in contrast, appears more contextual. Results for this group are
consistent with our initial hypothesis that giving Hispanic respondents a
chance to report Hispanic origin before asking race would reduce the extent of
"Other race" reporting.
These differences in race reporting for U. S.-born and migrant Hispanics
may reflect a process of acculturation. Our statistical results (namely, the
interaction effect between place of birth, form type, and reported race for
persons of Hispanic origin) imply that Hispanic respondents' assimilation and
use of the U. S. race category "White" are complex, and depend both on context
and on the category of person being described. When Hispanic identity has3
already been acknowledged (in the revised form, by the prior question about
Hispanic origin) then Hispanics (especially those born in the United States)
are more likely to report their race as "White." We should emphasize that we
offer these substantive interpretations quite tentatively; further testing on
larger and more representative samples is needed before they can be considered
more than hypotheses.
If they are replicated, our results imply that the measurement properties
of census items about race and Hispanic origin are affected by context
effects, and that the measurement properties of these items vary
systematically over the population. This result is consistent with Johnson's
(1987) finding that the measurement properties of different indicators of
-14-
Hispanicity are not constant, but vary between first or second generation
Hispanic immigrants versus others.
Discussion and Conclusions
Overall, these findings suggest that the revisions to the form tested in
these experiments could result in improvements in data quality. We increased
the response rate to the Hispanic origin item, which was one of the major
hypotheses guiding this research on question sequence. This was done without
affecting the substantive distribution, and without affecting the response
rate for the race item.
One limitation of the current study is the number of experimental
manipulations introduced in our revised questionnaire. Three experimental
manipulations could affect the results reported here: the order of the
Hispanic origin and race items, the added instruction to fill in the "No"
circle for Hispanic origin, and the variations in response categories for race
and Hispanic origin. A second limitation is the fact that results are based
on small and purposively selected "samples." We are presently planning and
conducting questionnaire experiments based on larger, randomly selected
samples, including a questionnaire experiment planned as an integral part of
the 1990 census.
Despite the limitations, our findings support some general conclusions
about questionnaire design. The first is that redundancy is a problem from
the point of view of respondents. Respondents apparently try to interpret
questions so as not to provide redundant information. When questionnaire
items really are partially or wholly redundant, a substantial amount of error
and missing data can result.
A second conclusion is that our results call into question the naive
assumptions about "error" with which we began our research. We aimed to
improve data quality by, among other things, reducing reporting by persons of
Hispanic origin in the "Other" category of the race item. As we analyzed our
-15-
data we came to the point of view that the patterns of reporting reflect real
differences in the interpretation of the meaning of race. We believe that the
reporting of Hispanics in the "Other" race category is not an error, but a
real perception by these respondents of their "correct" place in the racial
classification structure--neither "White" nor "Black." We agree with the
point of view stated by Lieberson (1985:160):
...some of the difficulties that researchers and census takersexperience in using data on racial and ethnic groups are duenot to problems of instrumentation or execution, pure andsimple--such as might occur if a question was constructedincorrectly through, say, some vagueness or ambiguity ofmeaning. Rather some of the difficulties and inconsistenciesreflect the processes of ethnic and racial change themselves;the `errors' are telling us something about the flux in theconcepts and identifications themselves.
Our findings concerning context effects on race reporting by Hispanics may
tell us something about the complex process of assimilation of U. S. racial
categories. Our speculation is that Hispanic respondents may have their own
"rules" for reporting race in the census for themselves and members of their
households, which do not conform to the Census Bureau's rules or the common
understanding of race. The rule (roughly stated) might be: apply the U. S.
racial category "White" to describe Hispanic persons with more "roots" in the
United States, but only after acknowledging Hispanic identity. If there is
such a rule, it has important implications for the meaning and the measurement
of race in the census and in general, which we plan to explore in our future
research on this subject.
-16-
REFERENCES
Bates, Nancy A., and Theresa J. DeMaio (1989)
"Using cognitive research methods to improve the design of the decennial
census form." Paper presented at the Census Bureau's Fifth Annual Research
Conference.
DeMaio, Theresa J., and Elizabeth Martin (1987)
"Documentation of revisions to 1986 test census long form." Unpublished
paper, U. S. Bureau of the Census. March 16, 1987.
DeMaio, Theresa J., Elizabeth A. Martin, and Betsy Page Sigman (l987)
"Improving the design of the decennial census questionnaire." Paper
presented at the 1987 Annual Meetings of the American Statistical
Association, Survey Research Methods Section.
Fay, Robert (1989)
"CPLX: Contingency Table Analysis for Complex Sample Designs, Program
Documentation." Unpublished, U. S. Bureau of the Census.
Fernandez, Edward W. and Arthur R. Cresce (l986)
"Who are the Other Spanish?" Paper Submitted for Presentation at the 1986
Annual Meeting of the Population Association of America, San Francisco,
California.
Harris, Marvin (1968)
"Race." In International Encyclopedia of the Social Sciences 13:263-268.
Johnson, Robert A. (1987)
"Evaluating U. S. Census Bureau measures of Hispanic origin: An approach
based on latent structure analysis." Paper presented at the Annual Meeting
of the American Statistical Association, San Francisco, California.
Lieberson, Stanley (1985)
"Unhyphenated Whites in the United States." Ethnic and Racial Studies:
8(1):159-180.
Marshall, Gloria (1968)
-17-
"Racial classifications: Popular and scientific." Pp. 149-64 in Science
and the Concept of Race, edited by M. Mead, T. Dobzhansky, E. Tobach, and
R. Light. New York: Columbia University Press.
McKenney, Nampeo R., Edward W. Fernandez, and Wilfred T. Masamura (1985)
"The quality of the race and Hispanic origin information reported in the
1980 Census." Proceedings of the Survey Research Methods Section (American
Statistical Association), pp. 46-50.
McKenney, Nampeo D., Arthur R. Cresce, and Patricia Johnson (1988)
"Development of race and ethnic items for the 1990 census." Paper
Presented at the 1988 Annual Meeting of the Population Association of
America, New Orleans, Louisiana.
Parsley, Theresa L., Theresa J. DeMaio, Elizabeth Page Sigman, Elizabeth
Martin, Pamela C. Campanelli, and Patricia Petrick (1989)
"Report on Results of the Classroom Testing for the Questionnaire Design
Project." 1986 Test Census, Preliminary Research and Evaluation Memorandum
No. 75. U. S. Bureau of the Census.
Passel, Jeffrey and David Word (1987)
"Problems in analyzing race and Hispanic origin data from the 1980 Census:
Solutions based on constructing consistent populations from micro-level
data." Paper presented at the 1987 Annual Meeting of the Population
Association of America, Chicago, Illinois.
Schuman, Howard and Stanley Presser (1981)
Questions and Answers in Attitude Surveys. New York: Academic Press.
Siegel, Jacob S. and Jeffrey S. Passel (1979)
"Coverage of the Hispanic population of the United States in the 1970
census: A methodological analysis." U. S. Bureau of the Census, Current
Population Reports, Series P-23, No. 82.
Turner, Charles F. and Elizabeth Martin (1984)
Surveying Subjective Phenomena. New York: Russell Sage Foundation/Basic
-18-
Books.
U. S. Bureau of the Census (1979)
Twenty Censuses: Population and Housing Questions, 1790-1980. Washington,
D.C.: U.S. Government Printing Office.
________ (l982)
Preliminary Evaluation of Responses in the Mexican Origin Category of the
Hispanic Origin Item. Persons of Hispanic Origin by State: 1980. 1980
Census of Population. Supplementary Report PC80-S1-7: 14-17.
________ (l986)
Content Reinterview Study: Accuracy of Data for Selected Population and
Housing Characteristics as Measured by Reinterview. 1980 Census of
Population and Housing Evaluation and Research Reports, PHC80-E2.
________ (1987a)
1990 Planning Conference Series No. 12: Race and Ethnic Items, July 1985.
Washington, D.C.: U.S. Bureau of the Census.
________ (1987b)
1986 National Content Test: Evaluation of Spanish/Hispanic Question
(Preliminary Draft). Evaluation report prepared by the Ethnic and Hispanic
Statistics Branch, Population Division. January, 1987.
-19-
FIGURE 1
Race and Hispanic Origin Items on the 1986 Form
FIGURE 2
Hispanic Origin and Race Items on the Revised Form
-20-
1. Strictly speaking, it is not quite accurate to say that race is based on
self-identification, since one respondent fills out the form for all household
members. This means that for most of the population, race (and all other
census information) is based on proxy reports. Nevertheless, it is the intent
of the Census Bureau to measure each person's racial self-identification, and
CONTEXT EFFECTS FOR CENSUS MEASURES
OF RACE AND HISPANIC ORIGIN
by
Elizabeth Martin, Theresa J. DeMaio, and Pamela C. Campanelli
Center for Survey Methods Research
U. S. Bureau of the Census
This paper reports the results of research undertaken by Census Bureau staff.A previous version was presented at the 1988 Annual Meetings of the AmericanStatistical Association. The views expressed are attributable to the authorsand do not necessarily reflect those of the Census Bureau. The authorsgratefully acknowledge the statistical assistance of Robert Fay and helpfulcomments by Robert Johnson and Irwin Schreiner. Any errors are the authors'.
ENDNOTES
-21-
the 1990 form includes a printed instruction to "Fill ONE circle for the race
that the person considers himself/herself to be."
2. Three American Indian respondents, each with large families, were randomly
assigned to receive the revised form, whereas only one Indian respondent,
living alone, was assigned to receive the 1986 form.
3. The reader may wonder how proxy vs. self reports of race are affected by
the form experiment. We do not have the data to examine this question
directly. However, in order to shed light on possible proxy effects, we
examined race-reporting in 38 Hispanic households in which one or both parents
and their children were living (nonrelatives and other relatives in the
household were ignored). In 92 percent of these cases, race was reported
consistently for the entire nuclear family, which suggests there are not large
proxy effects on race reporting.